The HYPERBOLA project is dedicated to estimating soil parameters and reconstructing layered scenes robustly, efficiently, and accurately. We explore algorithms applicable to both target and non-target scenes.
In target media, existing inversion methods based on imaging algorithms, such as back-projection (BP) and Stolt migration-based methods, encounter several issues: a) the Stolt migration-based method inherently depends on depth; b) the BP algorithm is inefficient due to the repetitive 'delay-summation' process required for each point in the imaging scene, particularly in layered media where refractive point calculations are necessary; c) these algorithms heavily rely on the judgement of the focus metric. To address these challenges, this project proposes a vertex extraction-guided local BP algorithm and improved refraction point calculation algorithms.
In non-target media, current inversion algorithms, like full-wave inversion and inverse scattering, struggle with low imaging efficiency, nonlinearity, and ill-conditioning. This project seeks to combine the strengths of full-wave inverse modeling and artificial neural networks to advance the capabilities of layered media reconstruction.